Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77369.70605237583,
-      "required_pages_per_second_upper_estimate": 0.005352615787154768
+      "required_token_positions_per_second": 76030.94560412878,
+      "required_pages_per_second_upper_estimate": 0.005259997232992825
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9671.213256546978,
-      "required_pages_per_second_upper_estimate": 0.004049135032618025
+      "required_token_positions_per_second": 9503.868200516097,
+      "required_pages_per_second_upper_estimate": 0.0039790711522200446
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19342.426513093957,
-      "required_pages_per_second_upper_estimate": 0.004028517348587979
+      "required_token_positions_per_second": 19007.736401032194,
+      "required_pages_per_second_upper_estimate": 0.003958810224617317
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29013.639769640937,
-      "required_pages_per_second_upper_estimate": 0.0017393234029271722
+      "required_token_positions_per_second": 28511.60460154829,
+      "required_pages_per_second_upper_estimate": 0.0017092271611633342
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29013.639769640937,
-      "required_pages_per_second_upper_estimate": 0.002255682874575271
+      "required_token_positions_per_second": 28511.60460154829,
+      "required_pages_per_second_upper_estimate": 0.0022166518484754002
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29013.639769640937,
-      "required_pages_per_second_upper_estimate": 0.3721558185456951
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28511.60460154829,
+      "required_pages_per_second_upper_estimate": 0.36571625045276857
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.04962567286719377,
-    "producer_p95_seconds": 0.147585835999962,
-    "gpu_update_seconds": 0.9529311116949242,
-    "producer_batches_per_second": 20.15086027500644,
-    "gpu_batches_per_second": 1.0493937995385176,
-    "producer_headroom": 19.202381683470975,
+    "producer_batch_seconds": 0.37138044991412045,
+    "producer_p95_seconds": 0.16997603199979494,
+    "gpu_update_seconds": 0.9697104174381896,
+    "producer_batches_per_second": 2.692656547298718,
+    "gpu_batches_per_second": 1.0312356988407223,
+    "producer_headroom": 2.611097104498714,
     "sustainable": true,
-    "buffer_seconds": 60.034660036780224,
+    "buffer_seconds": 60.12204588116776,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 470873.6991370569,
-    "upload_bytes_per_second": 5531121.898810259,
+    "generation_bytes_per_second": 417674.52390112757,
+    "upload_bytes_per_second": 5633587.339866974,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.03853445609836,
+    "estimated_full_shards_per_hour": 89.6232298638856,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -382,6 +382,12 @@
       "bytes": 6163668,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1962734589997126,
+      "bytes": 10535589,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 560900, "loss": 0.6532272100448608, "lr": 0.0003, "grad_norm": 0.225882472750135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82098261619, "tokens_per_second": 149525.24728713714, "data_wait_s": 1.5470999642275274e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561000, "loss": 0.6042928099632263, "lr": 0.0003, "grad_norm": 0.19715351121275068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82112786126, "tokens_per_second": 151525.81260820382, "data_wait_s": 1.548000000184402e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561100, "loss": 0.5654798150062561, "lr": 0.0003, "grad_norm": 0.15444098410477708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82127321036, "tokens_per_second": 151116.40383972894, "data_wait_s": 1.5919999896141235e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 561200, "loss": 0.6027804017066956, "lr": 0.0003, "grad_norm": 0.15055711764596116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82142153769, "tokens_per_second": 154513.84973032246, "data_wait_s": 1.771300048858393e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c534e40546345a5d948fb2717deb27d6f00e47ab70124ca18791d582a7fbd618
-size 85668
+oid sha256:1d10d64a494b42c11f1988c94fc9115585713085c21e1bf21b62dde2b2541fcd
+size 87786